Radiomics and AI tools transform uterine fibroid diagnosis

Uterine fibroids, or leiomyomas, present diagnostic challenges due to symptom overlap with rare but aggressive leiomyosarcomas. Standard imaging tools such as ultrasound and MRI often fail to distinguish these two conditions, leading to misdiagnosis or delayed treatment. AI-based systems are now narrowing this gap by learning complex patterns invisible to the human eye.

Artificial intelligence is rapidly redefining the diagnostic and therapeutic landscape of uterine fibroid management. With uterine fibroids affecting up to 77% of women globally, the rising demand for precision medicine has pushed innovation beyond conventional diagnostics and into the realm of AI-assisted healthcare. From aiding early detection and improving diagnostic confidence to customizing treatment plans and enhancing surgical outcomes, AI-driven solutions are proving indispensable.

A comprehensive new review titled “Artificial Intelligence and Uterine Fibroids: A Useful Combination for Diagnosis and Treatment,” published in the Journal of Clinical Medicine on May 15, 2025, presents the most in-depth synthesis yet of AI-enabled methods for fibroid management. Compiled by researchers across Europe and the Middle East, the review highlights the impact of deep learning, machine learning, radiomics, and haptic technologies on diagnostic accuracy and treatment outcomes.

How accurately can AI diagnose uterine fibroids and sarcomas?

Uterine fibroids, or leiomyomas, present diagnostic challenges due to symptom overlap with rare but aggressive leiomyosarcomas. Standard imaging tools such as ultrasound and MRI often fail to distinguish these two conditions, leading to misdiagnosis or delayed treatment. AI-based systems are now narrowing this gap by learning complex patterns invisible to the human eye.

Multiple approaches were reviewed. A deep convolutional neural network (DCNN) developed by Huo et al. significantly enhanced diagnostic precision in junior ultrasonographers, achieving over 94% accuracy and outperforming conventional methods. Meanwhile, a study by Chiappa et al. utilized radiomics on ultrasound data to distinguish fibroids from sarcomas, reaching 85% accuracy. Complementary research by Toyohara et al. used MRI and deep neural networks (DNN) to differentiate malignancies with 90.3% accuracy, comparable to senior radiologists.

Each approach offers distinct advantages: radiomics provides interpretability by isolating quantifiable image features, while deep learning systems autonomously detect subtle anomalies in large datasets. However, both approaches suffer from limited generalizability due to single-center datasets and rare sarcoma cases. The review calls for larger, multi-ethnic datasets and hybrid models combining interpretability and performance.

What are the AI advancements in surgical and non-invasive fibroid therapies?

Artificial intelligence is not only transforming diagnosis but also enhancing treatment precision. The review outlines how AI supports both traditional and non-invasive interventions - from hysteroscopic myomectomy to robotic-assisted surgeries and High-Intensity Focused Ultrasound (HIFU) ablation.

In surgical settings, Török et al. demonstrated that a fully convolutional neural network (FCNN) could identify fibroid tissue intraoperatively, enabling precise myoma-myometrium segmentation for real-time surgical guidance. Robot-assisted laparoscopic procedures also benefited from AI-powered 3D reconstruction tools, which allowed surgeons to localize and extract over 20 fibroids in a single session with minimal blood loss and postoperative complications.

HIFU procedures saw some of the most profound AI integrations. Akpinar et al. employed gradient boosting and other machine learning classifiers on MRI data to predict treatment success, achieving over 95% accuracy in identifying patients likely to benefit. Additional studies developed gadolinium-free monitoring techniques using deep learning, super-resolution DWI imaging for enhanced visualization, and automated 3D segmentation for treatment planning. These tools collectively advanced treatment personalization, minimized risks, and enabled real-time monitoring.

The most promising techniques integrated multimodal data, combining CE-T1WI and T2WI imaging with ensemble learning models to reach AUC scores near 0.90 in predicting HIFU ablation outcomes. Such models mark a leap toward AI-driven preoperative planning and intraoperative control.

How close is AI to clinical integration in gynecologic practice?

Despite significant promise, clinical adoption of AI in fibroid care faces several hurdles. The review highlights critical challenges: the “black box” nature of deep learning, lack of external validation, limited performance on rare or atypical presentations, and integration difficulties with existing hospital systems.

Ethical and regulatory issues also arise. Patient privacy, algorithmic bias, unclear accountability for AI-based decisions, and the absence of standardized reporting protocols impede clinical trust and deployment. Moreover, the economic implications of AI deployment in low-resource settings remain largely unexplored.

Nonetheless, hybrid diagnostic pathways integrating AI with traditional methods show strong potential. For example, haptic feedback systems developed by Doria et al. allowed robot-assisted surgeons to “feel” fibroids through a wearable display, restoring tactile sensation lost in minimally invasive procedures. Surgeons using this system achieved localization accuracy nearly equal to manual palpation, demonstrating how AI can replicate and enhance human expertise in real time.

In embolization therapies, deep learning models built on over 700 fibroid cases outperformed radiologists in predicting treatment success. These ensemble systems incorporated both imaging and clinical data, offering a holistic view for patient-specific interventions.